AnchorFlow: Training-Free 3D Editing via Latent Anchor-Aligned Flows

2025-11-27

AI Summary

Existing mask-free 3D editing methods suffer from timestep-dependent noise in diffusion sampling, limiting strong edit strength and geometric stability. To address this, we propose a global latent anchor mechanism that establishes a noise-invariant, consistent reference space. Our method employs anchor-alignment update rules and a relaxed alignment loss to achieve, for the first time without mask supervision, cross-trajectory-consistent, semantically precise editing. Built upon latent-space flow optimization, it jointly refines source and target trajectories while enforcing latent variable alignment. Evaluated on the Eval3DEdit benchmark, our approach significantly improves edit strength, structural stability, and geometric fidelity—enabling diverse, complex semantic transformations driven by natural-language instructions.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Learning with Manifolds

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Training-free 3D editing aims to modify 3D shapes based on human instructions without model finetuning. It plays a crucial role in 3D content creation. However, existing approaches often struggle to produce strong or geometrically stable edits, largely due to inconsistent latent anchors introduced by timestep-dependent noise during diffusion sampling. To address these limitations, we introduce AnchorFlow, which is built upon the principle of latent anchor consistency. Specifically, AnchorFlow establishes a global latent anchor shared between the source and target trajectories, and enforces coherence using a relaxed anchor-alignment loss together with an anchor-aligned update rule. This design ensures that transformations remain stable and semantically faithful throughout the editing process. By stabilizing the latent reference space, AnchorFlow enables more pronounced semantic modifications. Moreover, AnchorFlow is mask-free. Without mask supervision, it effectively preserves geometric fidelity. Experiments on the Eval3DEdit benchmark show that AnchorFlow consistently delivers semantically aligned and structurally robust edits across diverse editing types. Code is at https://github.com/ZhenglinZhou/AnchorFlow.

Problem

• Addresses inconsistent latent anchors in diffusion-based 3D editing
• Enhances geometric stability and semantic fidelity without fine-tuning
• Eliminates need for mask supervision while preserving shape integrity

Innovation

• Latent anchor consistency for stable 3D editing
• Mask-free approach preserving geometric fidelity
• Anchor-aligned update rule ensuring semantic faithfulness